Sampling is the process of drawing representative subsets of the population for
the purpose of measurement and inference. The key concept to take away is,
of course, that the sample must be representative of the population of interest
such that extrapolations can be made back to the larger group.
In reality, the Internet affords only a convenience sample of potential
respondents... there is no virtual counterpart to the random digit dialing
technique traditionally used in telephone interviewing. To that end,
online surveying may not be suitable for some research situations, such as
general consumer interest studies (see also the demographic weighting
discussion below).
Whatever our intentions are, a level of certainty must exist that the Internet
population is representative of the target population of interest.
Fortunately, by most recent accounts, the disparity and differences between
these two worlds are dissipating.
Samples can be drawn for online surveys in several ways:
Commercial Samples:
Statistical Reasoning does not have a vested interest in selling online survey
samples. In fact, we advocate using a qualified, internal sample database
where available. But should you still wish to obtain an online respondent
panel, SR can help you source an opt-in sample based on appropriate demographic
specifications, such
as gender, age, income, education, occupation, and the like.
Internal Customer/Member Samples:
In most project situations, the client company would have an internal sample
to be used in the study (e.g., a customer or employee or membership database).
Such scenarios are ideal since the sample is valid and is, by definition, 100%
representative of the population of interest. In these situations, emails
or intra-office memos could be used to contact the prospective respondents to
request their participation without worrying about SPAM and privacy concerns
(such as in the case of commercial samples).
Intercept Techniques:
Should neither an internal sample be available nor a commercial sample
affordable, intercept techniques in sample recruitment are possible.
Statistical Reasoning can set up executable scripts on partner websites that
spawn survey invitation messages when a respondent visits a particular page on
that website (e.g., the home page of a major search engine site). The
characteristics of the visitors of the website(s) of choice, of course, must be
consistent with the target population of interest for the particular study.
SR can help you determine whether that is indeed the case.
The survey invitation can be setup to open automatically when a respondent
visits a trigger page (i.e., PopOnEntry) or when a respondent leaves a trigger
page (i.e., PopOnExit). Additionally, the frequency in which these popup
windows appear can be determined a priori so as to obtain either a random or
systematic sample of website visitors.
A final alternative would be to have the invitation message as an element of a
webpage and the survey window open only when a link (agreement to participate)
is clicked (i.e., PopOnClick).
Click here to learn more on how Statistical
Reasoning can help you develop a meaningful and reliable survey instrument.  |
| :: POST- THEN PRE-TEST DESIGN :: | In a post- then pre-test design, survey participants could be recruited at the store front upon completing a purchase transaction. Participants would be informed of the company's customer satisfaction survey by the cashier and directed to the URL for the survey site (either printed on their transaction receipt or separate customer care card). An appropriate incentive for encouraging survey participation (e.g., discount on next purchase) may be suitable.
It is estimated that 800-1000 customers shop at the store per week. This group constitutes a reasonable sample frame size. For analytical/statistical purposes, a sample size of 384 would provide results accurate to within +/- 5%, 19 times out of 20. At the minimum, this should constitute the target number of completed surveys.
|
|  |  |
 |
| :: PRE-TEST, POST-TEST DESIGN :: | In a pre-test, post-test design, survey participants could be recruited at the store front upon completing a purchase transaction prior to the salesforce incentive program being implemented at that store. Participants would be informed of the company's customer satisfaction survey by the cashier and directed to the URL for the survey site (either printed on their transaction receipt or separate customer care card). An appropriate incentive for encouraging survey participation (e.g., discount on next purchase) may be suitable.
It is estimated that 800-1000 customers shop at the store per week. This group constitutes a reasonable sample frame size. For within-samples analytical/statistical procedures, a sample size of 250 would provide results accurate to within +/- 6.8%, 19 times out of 20. At the minimum, this should constitute the target number of respondents having completed both the pre- and post-test surveys.
|
| | |
|
| :: BETWEEN-SAMPLES DESIGN :: | In a between-samples design, survey participants could be recruited at the store front upon completing a purchase transaction prior to the salesforce incentive program being implemented at that store. Participants would be informed of the company's customer satisfaction survey by the cashier and directed to the URL for the survey site (either printed on their transaction receipt or separate customer care card). An appropriate incentive for encouraging survey participation (e.g., discount on next purchase) may be suitable.
It is estimated that 800-1000 customers shop at the store per week. Given that in a two discrete sample groups would be required in this design, the sample frame is estimated to be 1600-2000 (800-1000 x 2 stores) customers. This group constitutes a reasonable sample frame size. For between-samples analytical/statistical procedures, a sample size of 500, half of which are recruited from stores that have already implemented the incentive program and the other half from stores that have yet to implement, would provide results accurate to within +/- 6.8%, 19 times out of 20. This shall be the target number of completed surveys.
|
| | |
|
| ? |
:: MR INSIGHTS :: |
Sampling vs. Census:
Benefits of using a sample include:
1: Lower Costs
Regardless of which data collection method is used, the costs of acquiring each
additional response goes up proportionately. Albeit, with online surveys,
the incremental costs is significantly lower.
2: Faster Turnaround Period
Obviously, if there are fewer respondents to interview/survey, the data
collection phase of activities can be completed in a shorter period time,
thereby reducing field costs.
3: Better quality control
With fewer data records to clean, reduce, and manipulate, the chances for error
is reduced and a cleaner, more manageable data file can be delivered to the
statistical analysts.
4: Finite & Infinite Populations
At times, we may be dealing with a finite population (e.g., testing product durability
requires ‘breaking’ every unit) or infinite (e.g., impact of mosquito size on
repellant sales), it would be impractical, if not impossible, to conduct a
census. |
| | |
|
| @ |
:: INSIDE SR :: |
Demographic Weighting:
It has been proposed that propensity or demographic score adjustments (i.e.,
weighting) can be used to compensate for differences between demographic
characteristics of the online population relative to those not on online.
Given that most concerns regarding online research tend to revolve around the
sampling design and representativeness of online respondent populations, this
perspective is certainly appreciated. Caution, however, is warranted.
Click here to read more. |
| | |
|
|
Research Design
The Internet affords only a
convenience sample of potential
